Buyer Review Pack security and data handling
A record-only service designed to minimize access, isolate customer packages, and make issued output verifiable against separately published key material.
At a glance
| Control | Status | Verify it |
|---|---|---|
| Administrator accounts protected by hardware security keys | In place | Security page |
| Operator actions require a passkey and an allowlisted network | In place | Security page |
| Customer sign-in: single-use link, 15-minute expiry; sessions 30 min idle / 12 h | In place | Security page |
| Order database encrypted with a customer-managed key (Cloud KMS) | In place | Security page |
| Daily backups with point-in-time recovery; 30 retained; last restore test: 3 October 2026 | In place | Security page |
| Security logs kept 90 days; access to customer files logged | In place | Security page |
| Public site loads no third-party scripts; strict CSP, HSTS preload | In place | View headers |
| SPF, DKIM, DMARC on attestlayer.com | In place | DNS |
| 72-hour breach notification | Contractual | DPA §9 |
| Sub-processors published, change notice before processing | In place | Subprocessors |
| Vulnerability disclosure policy and security.txt | In place | security.txt |
| No security incident affecting customer data as of 2026-10-01 | Statement | Security page |
| Package signing keys in Secret Manager; public keys published with validity dates | In place | Registry |
Direct-client and Console scope
The standard Buyer Review Pack does not require access to customer production systems, installed endpoint software, production credentials, or source-code repositories. Customers submit only the records authorized for one eligible client review through the secure Console.
Accepted inputs and authorization
Client Console keeps the client request, any required completed response, and supporting source records in separate upload slots. Only the purchasing organization's verified, order-authorized session may access its order and package workflow. Customers must submit only records they are permitted to use for that client review.
Supported formats, size limits, hidden-content checks, archive limits, malware checks, and prohibited-content rules are enforced before generation. Do not upload credentials, secrets, private keys, payment-card data, health data, source code, production access, or unauthorized records.
Storage and encryption
The direct-client workflow uses Google Cloud resources in Montréal for application processing, private object storage, and the order database. HTTPS protects data in transit. The order database is encrypted at rest with a customer-managed key in Cloud KMS. Secrets and object storage use Google-managed encryption. Public access prevention, authenticated package access, tenant-scoped order checks, least-privilege service identities, and separately managed secrets constrain access.
Automated processing and review
AI points; it never writes. An AI model (Claude, by Anthropic) only selects sentences that already exist in your records. Every answer is a word-for-word quote, checked by fixed rules and by a second AI check, and nothing reaches your client until you approve it.
Generation is automated and may classify requirements, select source passages, draft source-linked responses, and identify gaps. It does not establish that a source record is true. The customer reviews every proposed answer and evidence decision before creating the final client-forwardable package. See the Automated Processing Notice.
L'IA pointe, elle n'écrit jamais. Un modèle d'IA (Claude, d'Anthropic) ne fait que sélectionner des phrases qui existent déjà dans vos documents. Chaque réponse est une citation mot pour mot, vérifiée par des règles fixes et par une seconde vérification par IA, et rien ne parvient à votre client avant votre approbation.
Pendant le lancement, le fondateur vérifie chacun des dix premiers dossiers avant leur émission ; cet accès est journalisé comme tout autre.
Retention and deletion
We keep a working copy of your request and records for 14 days after delivery so reruns work without re-uploading, then delete it. Generated package access follows the order window. Package objects, email challenges, session/Fit records, transaction metadata, backups, lawful holds, and deletion recovery have different lifecycle rules. The exact categories and periods are published in the Data Retention and Deletion Policy.
To check each question, we send only the relevant excerpts of your records and the question text, never whole files, to Anthropic's Claude API, which processes them in the United States. Under Anthropic's commercial terms, this data is not used to train models without permission (we have not given it) and is deleted from Anthropic's systems within 30 days, unless a usage-policy review requires longer retention. Anthropic is listed among our subprocessors.
Payments, email, and subprocessors
Stripe handles payment-card details. SendGrid handles transactional email. Google Cloud hosts the application and protected data path. No public analytics provider receives customer source records from this workflow. Current provider purposes and locations are listed on the Subprocessors page.
How access and keys are protected
Operator actions on the Control Desk require a passkey (phishing-resistant, bound to a registered credential) and are accepted only from an allowlisted network.
Customers sign in with a single-use email link that expires after 15 minutes. Sessions end after 30 minutes of inactivity or 12 hours at most. We store only hashes of the link and the session secret.
Package signing keys are held in Secret Manager; only the public keys are published, in the Registry, each with its key ID and validity dates.
La base de données des commandes est chiffrée au repos avec une clé gérée par le client dans Cloud KMS. Les secrets et le stockage d'objets utilisent le chiffrement géré par Google.
The order database is backed up daily (03:00 UTC) with point-in-time recovery; transaction logs are kept 7 days and 30 backups are retained.
Security and abuse-prevention logs are kept for up to 90 days. Access to customer files is logged.
Les comptes administrateur Google Cloud et Google Workspace sont protégés par des clés de sécurité matérielles.
Aucun incident de sécurité touchant des données clients n'est survenu en date du 2026-10-01.
Security and package-integrity practices
- HTTPS transport. The order database is encrypted at rest with a customer-managed key in Cloud KMS. Secrets and object storage use Google-managed encryption.
- Private object storage, public-access prevention, authenticated package access, and tenant-scoped authorization.
- Least-privilege service identities, separately managed secrets, restricted administrative access, and separation between public, checkout, Console, verification, and processing roles.
- Processing is automated. The founder may open delivered packages for quality control, under the same confidentiality obligations as the Service. No one else at AttestLayer has access to your records or packages.
- Business-email access verification, signed payment-webhook verification, replay and duplicate-event protection, and order-capacity reconciliation.
- Canonical manifests, SHA-256 file hashes, signed receipts, issuer-key separation, hosted and offline verification, and intentional-tamper rejection.
- Security logging, incident handling, controlled suspension, rollback, and recovery procedures.
Incident and vulnerability contact
Report a suspected product-security incident or data-handling concern to security@attestlayer.com. Do not include secrets or confidential source records in ordinary email. Security researchers should follow the Vulnerability Disclosure Policy.
Current assurance
AttestLayer does not currently have:
- SOC 2 (Type I or II), ISO/IEC 27001 or CSA STAR;
- an independent penetration test or external assurance opinion;
- cyber or technology errors-and-omissions insurance.
AttestLayer does not currently hold cyber or professional liability insurance. The liability limit in the Terms of Service applies.
AI points to your own sentences. It never writes answers.
AttestLayer does not currently claim or display completion badges for SOC 2, ISO/IEC 27001, an independent penetration test, or insurance.
A PACKAGE INTEGRITY VERIFIED result confirms the package checks identified by the verifier. Buyers can repeat that check with the hosted verifier. It does not certify compliance, establish that submitted records are true, or guarantee client acceptance.
Public site
The public site loads no third-party scripts.
Where records are processed
Customer questionnaires and records are stored in Montréal (northamerica-northeast1). Only relevant excerpts and question text are processed by Anthropic in the United States. Business email addresses go to SendGrid. Card and invoice data go to Stripe. See the Data Processing Addendum for the 72-hour breach notice.
DPA and further questions
The direct-client Data Processing Addendum and the policies linked above describe the standard public terms. Email security@attestlayer.com to request further security material; an NDA is available. Completed answers about AttestLayer are on the security questionnaire.